How to Create Amazon Product Photos in Claude With an MCP Connector

Key Takeaways

• Claude doesn't generate images itself; a connector lets it drive a tool that does. MCP lets Claude call an image tool for you, so you describe the shots in chat and the tool renders them.

• Setup is a few clicks. In Claude, go to Customize, then Connectors, add the tool's server URL or pick it from the directory, and sign in. Claude's free plan allows one custom connector.

• Start with Amazon's main-image rules. Pure white background (RGB 255, 255, 255), the product filling 85% of the frame, no text, and at least 1,000 px on the longest side if you want zoom.

• Ask for the main image first, then the secondary set in the same chat. Claude keeps the context, so one product description carries through every shot.

• Check every output against the real product. AI edits can shift colors and redraw logos or label text, and Amazon wants the main image to show the product accurately.

Why make Amazon product photos in Claude

If you sell on Amazon, you need the same set of images for every product: a white-background main image, a few angles, a lifestyle shot or two, and something that explains the features. Doing that in Photoshop for each SKU is slow, and most AI image apps make you click through the same steps one product at a time.

A connector changes the workflow. MCP (Model Context Protocol) is an open standard that lets Claude use outside tools. Hook up an image tool and you can ask for "a white-background main image and three lifestyle shots on a kitchen counter" in plain English, and Claude handles the tool calls. It's the same tool you'd use in its own app, driven from a conversation instead.

Step 1: Know what Amazon will accept

Amazon's product image guide is strict about the main image, the one shoppers see in search results. It needs a pure white background (RGB 255, 255, 255), the product should fill 85% of the frame, and it can't carry text, badges or props that don't come with the product. Get it wrong and the listing can be suppressed from search.

The technical rules are easy to hit. JPEG, PNG, TIFF or non-animated GIF, RGB color, and between 500 and 10,000 pixels on the longest side. Aim for at least 1,000 px, because that's what turns on zoom. Secondary images are looser: other angles, the product in use and close-up details, and Amazon recommends at least six of them plus a video. Plenty of sellers add text callouts there; Amazon's guidance says any text has to be in the marketplace's language or English.

There's one AI-specific rule. If an image includes a photorealistic person generated entirely by AI, Amazon wants it tagged before upload so it can add a disclosure. Amazon doesn't ban AI-made product images outright. The test is whether the image shows the product accurately.

Step 2: Add the connector to Claude

Pick a tool from the overview at the end of this post and grab its server URL. Then, in Claude on the web or the desktop app, open Customize, then Connectors. Click the + button, choose Add custom connector, give it a name, paste the URL and click Add. Then connect it: Claude opens the tool's sign-in page, you log in and approve access. Anthropic's custom connector guide covers the details.

Some tools, like Adobe and Canva, are listed in Claude's connector directory instead. Click the + next to Connectors to browse it, find the tool, connect it and sign in. On a Team or Enterprise plan, an owner has to add the connector for the organization first.

Claude's free plan allows one custom connector. Pro ($20 a month, or $17 a month billed annually) doesn't have that limit. Some image tools also want a paid Claude plan, so check the tool's own setup page.

Step 3: Turn it on and hand over your product photo

In a new chat, click the + button, open Connectors and switch the tool on. The first time Claude calls it, you'll be asked to allow the action once or always.

Start from a real photo of the product, shot on any plain surface in good light. Getting it to the tool is the part that varies. Several image servers want an image URL or their own upload step rather than a file you drop into the chat, so check the tool's docs. Mage and Melius both have an upload tool Claude can call, and a link to a photo you've already hosted usually works everywhere.

Step 4: Ask for the main image first

Be specific and give Claude Amazon's rules up front, because it will pass them to the tool. A prompt that works:

"Using the attached product photo, remove the background and put the product on pure white (RGB 255, 255, 255). Keep the product, its colors and the label text exactly as they are. Center it so it fills about 85% of the frame. Export a square JPEG at 2,000 by 2,000 px."

Once that's right, ask for the rest in the same chat: "Now make three secondary images: the product in use on a kitchen counter, a close-up of the lid, and a version with three short feature callouts in large text." Claude keeps the context, so you don't need to re-explain the product.

Step 5: Check, name and upload

Look at every image next to the real product before it goes anywhere near Seller Central. AI edits tend to fail in the same places: colors drift a shade, small text on labels gets redrawn into nonsense, and logos come back slightly wrong. Any of those on the main image is a returns problem waiting to happen.

Name files the way Amazon expects, such as B000123456.MAIN.jpg for the main image and B000123456.PT01.jpg for the first secondary, or use Seller Central's flexible naming and assign them after upload. Images can take up to 24 hours to show on the listing.

A few things to know before you start

• You pay the image tool, not just Claude. Most connectors bill per image or in credits from your account with that tool, on top of your Claude plan.

• Some connectors need a paid Claude plan. Photoroom's does, for example. Check the tool's setup page before you pick one.

• Read the output rights. Most of these tools let you use images commercially, but the terms differ. Mage, for one, puts output in the public domain, so you don't get exclusive rights.

• Keep your original photo. If an edit goes wrong, you'll want the untouched shot to start again from, and it's your proof of what the product really looks like.

Tools you can connect

These all have an official MCP server or Claude connector, checked against each vendor's docs and pricing on October 6, 2026. They're listed alphabetically.

• Adobe: Adobe's connector brings 50-plus Photoshop, Lightroom, Firefly and Express tools into Claude, including Generative Fill and Replace Background. About 40 work without an Adobe account. Find it in Claude's connector directory.

• Canva: Built for layouts, so it's useful for the text-heavy secondary images like feature callouts and size charts. Filling brand templates needs Canva Enterprise. It's in the connector directory.

• fal.ai: Its MCP opens fal's whole model catalog to Claude. The connector is free and you pay per model run. Server: https://mcp.fal.ai/mcp-relay.

• Mage: Mage's MCP server pairs with a free skills pack that includes a marketplace-cards skill for a white-background main image, secondary images and A+ modules. Its API is labeled beta, and generations are paid in Gems, with plans from $10 a month. Server: https://mcp.mage.space/mcp.

• Melius: A canvas-based creative platform whose MCP server lets Claude remove backgrounds, resize to several aspect ratios and download a whole set as a ZIP, which suits sellers repeating one look across many SKUs. Plans start at $20 a month. Server: https://mcp.melius.com/mcp.

• Photoroom: A product-photo editor whose connector handles background removal, shadows, relighting and upscaling, the core jobs for a main image. Background removal is $0.02 an image through its API. Server: https://mcp.photoroom.com/mcp.

What this changes for Amazon sellers

The image tools themselves aren't new. What a connector removes is the clicking: the same product goes through the same edits for the main image, the angles and the infographics, and you describe it once in a chat instead of rebuilding it in an app for every SKU. Start with one product, get the main image passing Amazon's rules, and only then scale it to the rest of the catalog.